부산대학교 도서관

Global gene expression analysis has been a major tool for urothelial carcinoma subtype discovery. This approach has revealed extensive complexity both in intrinsic features of the tumor cells and in the microenvironment. However, global gene expression cannot distinguish between gene expression signals originating from the tumor cells proper and from normal cells in the biopsy. Here, we use a large cohort of advanced urothelial carcinomas for which both gene expression data and extensive immunohistochemistry are available to create a supervised mRNA expression centroid classifier. This classifier identifies the major Lund taxonomy tumor cell phenotypes as defined by IHC. We apply this classifier to the independent TCGA dataset and show excellent associations between identified subtypes and genomic features. We validate a progressed version of Urothelial-like A (UroA-Prog) that shows FGFR3 mutations and CDKN2A deletions, and we show that the variant Urothelial-like C is almost devoid of FGFR3 mutations. We show that Genomically Unstable tumors are very distinct from Urothelial-like tumors at the genomic level, and that tumors classified as Basal/SCC-like all complied with the established definition for Basal/SCC-like tumors. We identify the Mesenchymal-like and Small-cell/Neuroendocrine-like subtypes, and demonstrate that patients with UroB and Sc/NE-like tumors show the worst overall survival.